The Cupertino giant has unveiled VoxelNet. This is not the code name of his future smartphone. No. This is the name of the method developed by Apple and described by two of its researchers in a scientific article published November 17 last.
An original approach
VoxelNet is at the heart of the operation of an autonomous car since it allows the vehicle to identify and locate cars, cyclists, pedestrians, etc. thanks to a LiDAR. A rotating camera on all the autonomous cars developed by Apple’s competitors and which emits laser beams at 360 °. The latter will then bounce on the various obstacles and return to the LiDAR which can then determine the position and distance of the objects in question according to the time taken by the round trip. From this data received by the LiDAR, an embedded computer creates a cloud of points which are then processed to determine if the path is free.
Where Apple’s method differs is that its engineers use a neural network to save time and delete a stage. The data extraction and the box prediction are grouped together. To be more precise, “VoxelNet divides a cloud of points into 3D voxels (a set of 3D pixels, so cubes of sorts, Ed) spaced equally. It then transforms a group of points within each voxel into a unified representation. “This approach makes it possible to describe the cloud of points “as a detailed volumetric representation” of the environment which is then submitted to the neural network in charge of detecting objects (cars, pedestrians, etc.).
According to the Apple article, which has not yet been approved by peers, this approach is much more efficient than the best current 3D detection methods using LiDAR.
The weight of the culture of secrecy
After last summer’s confirmation by Tim Cook of Apple’s interest in autonomous cars, this publication proves that the US giant’s engineers are making progress and finding innovative ways to compensate for their company’s lag in the industry.
It may also be the surface manifestation of a frenetic occupation under the clevis of secrecy that Apple maintains around all its activities. A policy of absolute silence that goes against his interest sometimes. It helps to give the impression that Tim Cook’s company is lagging behind the competition, Google, Facebook or Microsoft who publish many scientific reports each year. Worse, it does not facilitate the recruitment of top talent in academic spheres, even though Apple’s boss of artificial intelligence research, Ruslan Salakhutdinov, continues to publish regularly as part of his chair at Carnegie Mellon University. In the field of artificial intelligence, the Cupertino Company has, a priori, published only three articles for the time being. The first was a landmark. To the point of being awarded one of the two best scientific article awards at the very annual Computer Vision and Pattern Recognition conference.
If for the moment autonomous cars are the first concerned by this publication, the detection of 3D objects using point clouds can also be used in other areas. The article by Apple researchers lists particular virtual and augmented realities among the sectors that can benefit from it.
But Tim Cook has more than once signified his company’s interest in augmented reality. A recent rumor even wants that Apple is actively working on an AR helmet that could be marketed as early as 2020.
We know that Apple is adept at spreading its technologies through its product lines. From there, to imagine that the progress made by his teams in the field of autonomous cars are tomorrow in our smartphones and augmented reality glasses…